Building Multi-Layer, Multi-Year Geospatial Data

SystemEarth · Data Engineering and Environmental Science · Website

The problem

SystemEarth’s evaluation of geospatial risk for farm plots is a composite of multiple geospatial layers, such as tree cover loss, natural forest baselines and restricted use areas, among others. Yet, a compliance decision rarely needs the entire raw value of data, tending to require segments of pixel information: presence of a forest at a particular plot of land at the regulatory cut-off date, degradation over time, overlapping geometry, social, environmental and governance risks etc, creating a challenge of handling vast amounts of information in a performant and accurate manner. Problem statement: Present on a specific use case and demonstrate how SystemEarth can represent multiple geospatial layers across time and space in a compact and query-efficient form, while preserving the information required to reproduce plot-level compliance decisions.

What you produce

- Analysis of which layers and attributes are essential for decision making, whether any of them can be simplified, summarised or dropped. This should be grounded in regulatory definitions for the specific use case selected, e.g. forest thresholds and cut-off dates. - Proposed encoding strategy combining multiple layers and time steps into a single dataset or architecture. - Working prototype that encodes a sample set of layers for one region, and queries against the compressed data. - Evaluation comparing the proposed approach against the original layers across: - Storage efficiency: reduction in storage requirements. - Query performance: query speed (effort, processing time) and resource requirements (additional storage) for representative plot-level queries. - Preservation of plot-level risk results, evaluated against: - Decision fidelity: percentage of plots for which all compliance-relevant outputs from the compressed representation match those derived from the original layers. - Pixel fidelity: degree to which the encoded representation preserves the original pixel-level information, where pixel-level fidelity is relevant to the use case. - Bonus: Explore the estimated infrastructure and processing costs associated with storing, updating and querying the proposed representation compared with the original layers.

The commitment

Starts

Sun 11 Oct, 00:00 SGT

Ends

Sat 17 Oct, 03:00 SGT

Time

On site with the company. Enter the room access code below to apply; submissions open through the end time.

You get

A certificate, judge-attested skills on your profile, and the work itself to show